ArticlePDF Available

Instrument Development for the FocaL Adult Gambling Screen (FLAGS-EGM): A Measurement of Risk and Problem Gambling Associated with Electronic Gambling Machines


Abstract and Figures

Previous research, based on a survey of 374 electronic machine gamblers living in Ontario, Canada, led to the selection of statements and the creation of ten constructs for the development of a new instrument, the FocaL Adult Gambling Screen for Electronic Gambling Machines (FLAGS-EGM). In this study, we used the Partial Least Squares Path Analysis form of Structural Equation Modelling to produce a hierarchical set of the ten constructs with proven predictive power for problem gambling. Receiver Operating Characteristic analysis identified cut off values for all of the constructs that predicted the target values with the desired degree of accuracy. Active gamblers were placed in five categories: No Detectable Risk, Early Risk, Intermediate Risk, Advanced Risk and Problem Gamblers. As described here, the FLAGS-EGM instrument has the potential to be applied in many situations in which identification of at-risk EGM gamblers is needed. © 2015 Centre for Addiction and Mental Health. All rights reserved.
Content may be subject to copyright.
Instrument Development for the FocaL Adult Gambling
Screen (FLAGS-EGM): A Measurement of Risk and Problem
Gambling Associated with Electronic Gambling Machines
Tony Schellinck,
Tracy Schrans,
Heather Schellinck,
& Michael Bliemel
Dalhousie University, Halifax, Nova Scotia, Canada.
Focal Research Consultants Limited, Halifax, Nova Scotia, Canada.
Previous research, based on a survey of 374 electronic machine gamblers living
in Ontario, Canada, led to the selection of statements and the creation of
ten constructs for the development of a new instrument, the FocaL Adult
Gambling Screen for Electronic Gambling Machines (FLAGS-EGM). In this
study, we used the Partial Least Squares Path Analysis form of Structural
Equation Modelling to produce a hierarchical set of the ten constructs with proven
predictive power for problem gambling. Receiver Operating Characteristic
analysis identied cut off values for all of the constructs that predicted the target
values with the desired degree of accuracy. Active gamblers were placed in ve
categories: No Detectable Risk, Early Risk, Intermediate Risk, Advanced Risk
and Problem Gamblers. As described here, the FLAGS-EGM instrument has the
potential to be applied in many situations in which identication of at-risk EGM
gamblers is needed.
Des recherches fondées sur une enquête menée auprès de 374 joueurs de jeux de
hasard électronique ont conduit à la sélection dénoncés et à la création de dix
construits destinés à la mise au point dun nouvel instrument appelé FocaL Adult
Gambling Screen for Electronic Gambling Machines (FLAGS-EGM). Nous avons eu
recours à une analyse des pistes causales par la technique des moindres carrés, une
forme danalyse des équations structurelles, en vue de produire un ensemble
hiérarchique constitué des dix constructs ayant démontré une efcacité prédictive
relativement aux problèmes de jeu. Une analyse de la fonction defcacité du
récepteur a permis de dénir des valeurs seuils pour tous les constructs ayant prédit
des valeurs cibles avec le degré de précision anticipé. Les joueurs actifs ont été
répartis en cinq catégories : aucun risque détectable, risque précoce, risque
intermédiaire, risque accru et joueur compulsif. Linstrument FLAGS-EGM
Journal of Gambling Issues
Issue 30, May 2015 DOI:
pourrait sappliquer à un grand nombre de situations où il est nécessaire didentier
les joueurs à risque parmi ceux qui sadonnent aux jeux de hasard électronique.
Few gambling assessment screens have been specically designed to identify an
individuals risk for harmful consequences prior to the onset of actual problems. It
was our objective, in developing the FLAGS-EGM, to create such an instrument.
The FLAGS-EGM also categorized individuals as problem gamblers, although this
was not in fact its main purpose. We also wanted to design a screen that could be
self-administered. During the extensive development phase of this research
(Schellinck, T., Schrans, Schellinck, H., & Bliemel, in press) we focused on ensuring
the statements considered for inclusion in the instrument were clearly understood
and consistently interpreted by gamblers (Appendix). Ideally, this measure would
educate and alert individuals regarding the likelihood of their risk of becoming
problem gamblers and motivate them to adopt behaviours that would reduce their
chances of experiencing harms.
Maddern and Rogala (2006) administered a 36-statement pilot version of the
instrument to a sample of at-risk gamblers. These individuals found the statements
were easy to understand and were an accurate assessment of their beliefs and
behaviours. Many subjects indicated that it would motivate them to change their
behaviours in regards to gambling. Buckley (2013) found that administering the
FLAGS-EGM to a sample of gamblers and then providing them with their
indicators of risk and classication of risk signicantly increased their readiness to
change their gambling behaviour.
We tested the validity and reliability of the ve reective and ve formative constructs in
the FLAGS-EGM (Table 1) as a further step in developing the instrument (Schellinck
T. et al., in press). In the current study, we modelled the relationships among these
constructs to determine the nature and timing of their inuence in the evolution of a
problem gambler. Constructs found to be signicantly positioned along the path to
problem gambling were used to create indicators of risk. If any of the variables were
hierarchical in nature, those variables found to precede other constructs were considered
to be earlier indicators of risk. Using Receiver Operating Characteristic (ROC) analysis
(Metz, 2006) we examined the predictive nature of the constructs to establish cut offs
such that individuals who scored at or above the designated level would be considered at
risk. Those constructs found to be directly connected to the problem gambling
constructs were considered indicative of the most advanced level of risk.
As described in Schellinck, T. et al. (in press) the new instrument was based on a
research model previously created to identify antecedents of problem gambling as
well as on an extensive review of the literature. To create the instrument structure
and scoring system we needed to complete the following steps:
Demonstrate that the construct scores were related to harms due to gambling,
Establish a hierarchy to the constructs in terms of when they would be manifested
prior to the gambler becoming a problem gambler,
Determine sum score levels for each construct that would accurately provide an
indication that a person can be characterised as at risk by this construct, and
Set criteria by which these indications would assign gamblers to various levels of risk.
Partial Least Squares Path Analysis form of Structural Equation Modelling (PLS-
SEM) (Chin, 1998; Chin & Newsted, 1999; Hair, Ringle, & Sarstedt, 2011) was used
to achieve steps 1, 2 and 4. We chose this method as PLS-SEM has become the
common one of investigating in the area of management research the cause-effect
relations between latent constructs. It maximises the explained variance of the
dependent latent constructs similar to multiple regression analysis. In particular,
when the goal of the model development process is prediction and theory
development, as it was here, PLS-SEM is viewed as the most appropriate method
of analysis (Hair et al., 2011). PLS-SEM provides many advantages over standard
Structural Equation Modelling. It identies key driver constructs when predicating a
target construct, can easily accommodate both formative and reective constructs, is
used for exploratory research into structural theory, and is suitable for use in a
complex structural model with many constructs and indicators. Moreover, PLS-
SEM can be used with a relatively small sample. The analysis is built on the
properties of Ordinary Least Squares (OLS) regression which means that traditional
methods of estimating the sample power as outlined by Cohen (1992) can be used
(Hair, Hult, Ringle, & Sarstedt, 2013). Our largest construct (Negative Con-
sequences) had 14 items which, extrapolating from gures presented by Hair et al.
(2013, p. 21), indicated that a sample size of 293 or larger would provide a power of
80% or better, a minimum R
of 0.1 and a 1% signicance level. Our sample size of
374 was clearly adequate for the required analysis.
Table 1
Construct Type and Number of Statements in each FLAGS-EGM Construct
Construct Construct Type Number of Statements
Erroneous Cognitions Beliefs Formative 5
Erroneous Cognitions Motives Formative 4
Preoccupation Desire Reective 4
Preoccupation Obsession Reective 2
Risky Behaviours Earlier Formative 6
Risky Behaviours Later Formative 6
Impaired Control Continue Play Reective 5
Impaired Control Begin Play Reective 3
Negative Consequences Formative 14
Persistence Reective 4
FLAGS-EGM Instrument (Beta): Total Statements 53
To establish accurate sum score levels for each construct (step 3, above) ROC
analysis, using each construct as a predictor of an appropriate target variable, was
required. In this case, ROC analysis was used to assign cut off points on the summed
scores to provide indicators of risk for respondents. Combinations of these indicators
were used to allocate electronic machine gamblers to risk categories.
Information obtained from a sample of regular EGM gamblers who played ‘‘ the slots’’
on average at least once a month over the previous year was used to develop the
instrument. Over a ve-day period, potential respondents were asked to participate in a
research panel as they entered a casino in Ontario, Canada. Telephone interviews were
conducted in April and May of 2009 with the sample of panel members. A total of 422
surveys were completed out of 610 eligible panel members (69.2%), with 48 disqualied
because of respondent selection criteria (e.g., played slots less than once per month over
the previous year), leaving 374 completed surveys available for analysis. This sample
comprised 150 males (40.1%) and 224 females (59.9%); the median age was 63 with ages
ranging from 23 to 89. The rst language of 85.5% of the participants was English, 46.8%
were retired and 2.1% were unemployed.
The participants indicated they had never received any treatment or services for substance
use or gambling or mental health issues. Slightly over half of participants (53.5%)
indicated they gambled weekly or daily on the slots. 70.1% also purchased lottery tickets
at least once a month, 2.4% had participated in Internet gambling in the last year, 2.1%
played casino table games monthly, 10.6% played card games for money monthly, 11.5%
went played bingo in bingo halls monthly, and 7.2% gambled on horse racing monthly.
The survey comprised 132 dichotomous statements that were randomized for each
participant to reduce the risk of common method bias (Bliemel & Hassanein, 2007).
The survey also gathered demographic information, general gambling behaviour and
playing patterns. Briey, the statements were formed into a set of ve formative and
ve reective constructs, with 53 statements, for inclusion in the FLAGS-EGM
instrument. The specic process through which the statements were formed, and the
logic underlying the process, is described in Schellinck, T. et al. (in press).
PLS-SEM Analysis
Four criteria were established by Urbach and Ahlemann (2010) that can be used to
evaluate the validity of the SEM-PLS model derived using the ten constructs:
Coefcients of determination (R
) where values of 0.670, 0.333 and 0.190 were
considered substantial, moderate and weak respectively.
Signicant path coefcients using bootstrapping with 5000 runs.
Independent latent variables having substantial impact on dependent latent values (f
with values of 0.35, 0.15 and 0.02 considered to be large, medium and low effect levels.
Predictive Relevance (Q
) where the threshold for signicant impact was 40.
Development of Risk and Problem Gambling Categories
To classify gamblers into risk categories we grouped the constructs based on the
following six criteria: 1. The constructs were ordered based on the hypothesized
direction of causality conrmed by the strongest predictive relationships found in the
PLS-SEM model. 2. Constructs in the PLS-SEM model needed to be directly
connected to constructs in the next highest level of risk/problem gambler. 3. The
latter constructs needed to have a major impact (f
) on the higher risk/problem
gambler constructs. 4. Higher risk/problem gambler constructs should be inuenced
by lower risk constructs. 5. Cognitive-based constructssuch as risky beliefs and
motives, when placed at the beginning of the PLS-SEM modelwere designated
early indicators and grouped accordingly. 6. If a gambler exhibited behavioural
based indications (i.e., Impaired Control and Risky Practices) they were classied at
a more advanced risk level (i.e., Intermediate or Advanced Risk levels). Once
gamblers were engaged in risky behaviours they were considered to be at a more
advanced stage in the progression towards becoming a Problem Gambler.
It should be noted that three of the constructs were split into two parts during the
construct development phase of the study: Preoccupation Desire and Preoccupation
Obsession, Impaired Control Continue and Impaired Control Begin, and Risky
Practices Earlier and Risky Practices Later (Schellinck, T. et al., in press). Each resulting
pair had an earlier and later risk construct that could be inferred based on the frequency
of responses to the statements and their ultimate positioning in the PLS-SEM model.
Consequently, the decision was made to place those gamblers who had an indication of
risk on the later risk constructs into the Advanced Risk category.
ROC Analysis
The two criteria used to assign indicator cut offs for each of the ten constructs were
sensitivity (true positive rate) and 1 minus (-) specicity (false negative rate) as determined
by ROC analysis (Metz, 2006). The accuracy of a predictor variable or model, in
correctly classifying a person (the target value in the target dichotomous state variable),
could be assessed over the range of the predictor variables values. For each possible
value of the predictor variable a classication matrix was produced and the sensitivity,
specicity and the chi-square statistic for the matrix calculated. ROC analysis used these
values to produce a graph of the ROC curve based on sensitivity and 1- specicity such
that the power of the model to classify gamblers could be assessed visually.
As a diagonal line in the graph indicates a performance level no better than chance, the
greater the separation of the ROC curve from the diagonal the better the models
performance. The degree of separation is summarized by the total area under the curve;
the closer to 100% area coverage under the curve, the better the model performance. The
ROC analysis also produces an overall signicance test. As Conigrave, Hall, and
Saunders (1995) recommended, the predictor variable indicator value was selected at the
point in the ROC curve that corresponded with the maximized chi-square test score.
This approach weighted the sensitivity and specicity equally.
The state variable used depended on the construct being evaluated. As FLAGS-
EGM is a hierarchical model, we did not expect the constructs at the beginning of the
hierarchy (i.e., Risky Beliefs and Motives) to predict the target value at the end of the
hierarchy (i.e., problem gambling) accurately. Negative Consequences and
Persistence were evaluated using a score of 8+on the PGSI as the state value.
The PGSI Problem Gambler category was chosen as the value because it is
commonly used to identify problem gamblers and because it has been shown to have
considerable convergent validity with other instruments, such as the DSM-IV (Ferris
& Wynne, 2001). The other constructs were evaluated using the indicators in the
higher levels of risk as the target values. Specically, the FLAGS-EGM Problem
Gambler indicator was used as the target variable for Preoccupation Obsession,
Impaired Control Begin and Risky Practices Later. The FLAGS Advanced Risk
indicator was used for Risky Practices Earlier and Impaired Control Continue. The
FLAGS-EGM Intermediate Risk indicator was used for Preoccupation Desire,
Risky Cognitions Motives and Risky Cognitions Beliefs.
Comparison to PGSI
A modied version of the Problem Gambling Severity Index (PGSI) component of
the Canadian Problem Gambling Index (Ferris & Wynne, 2001) was administered
during the interview to provide a measure of problem gambling status and to assess
concordance in categorizing EGM gamblers as at risk or Problem Gamblers with the
new instrument. The statements in the PGSI were modied (see Table 8 for the
modied statements) by specically referencing slot play and casino play as the form
of gambling indicated in the statements. This change ensured that the two
instruments would be referencing the same behaviour when it came to identifying
sources of the risk or problem gambling status. To compare the success of PGSI and
FLAGS-EGM in this context we created two dichotomous variables that identied
problem gamblers in each of the instruments and which produced a tetrachoric
correlation as a measure of concordance. The tetrachoric correlation is considered
the appropriate statistic, rather than Pearson or Spearman correlations or the kappa
statistic, when comparing two categories (Bonett & Price, 2005; Uebersax, 1987).
As the PGSI has two risk categories and the FLAGS-EGM has three, we needed to
combine two of the FLAGS-EGM categories together to compare the risk
classications between the two instruments. The PGSI uses a fairly wide range of
scores (37) to assign gamblers to its Medium Risk category. Consequently, for
comparison purposes, we combined the Intermediate and Advanced Risk categories
of the FLAGS-EGM into a single category equivalent to the PGSI Medium Risk
category. As some of the gamblers could fall into either the No Detectable Risk/No Risk
or the Problem Gambler categories at either end of the scale for both instruments, in this
instance it would not be appropriate to use the tetrachoric correlation. We measured
overlap by taking the sample of all gamblers identied as at risk by either instrument
and determined the percent of common assignment to risk level.
To aid in interpreting any discrepancies found between the classication by the two
instruments we created four discrepancy segments, PGSI at Low Risk or Medium Risk
but FLAGS-EGM No Detectable Risk, PGSI No Risk but FLAGS-EGM at Early
Risk or higher, PGSI Low Risk but FLAGS-EGM Intermediate Risk or higher, and
PGSI Medium Risk and FLAGS-EGM Problem Gamblers. Each segment was
compared on the ten FLAGS-EGM constructs and the nine PGSI statements. Author
judgment was used to interpret the results.
Risk Levels Based on Partial Least Squares Analysis
Using SmartPLS the analysis started with a saturated model using all ten constructs,
all constructs connected, and then, non-signicant paths removed. The direction of
the signicant paths was then reversed, one path at a time, to ensure the largest
coefcients occurred when the path was in the expected direction. The resulting
eighteen paths were all signicant at the p o0.05 level based on t-scores derived
from 5,000 bootstrapping runs (Figure 1).
Figure 1. PLS Model Showing Path Coefcients, T-Scores and Variance Explained in
Each Construct
The variance explained (R
) was .633 for Negative Consequences and .718 for
Persistence, which puts the model into the substantial variance explained range.
Similarly, the variance explained for Risky Practices Earlier and Risky Practices
Later was .636 and .613 respectively.
The relative effect size (f
) for each preceding construct on the target construct (listed
at the top of the column) (Chin, 1998) is presented in Table 2. All of the effect levels
were above, or near, the medium level of 0.15 suggested by Urbach and Ahlemann
(2010). Risky Practices Later was strongly inuenced by Risky Cognitions Motives
(0.47), while Negative Consequences was strongly inuenced by Impaired Control
Begin (0.54) and Risky Practices Later (0.49). Negative Consequences (0.57) had the
most effect on Persistence. All ten constructs met the criterion for predictive
relevance (Q
) of values greater than zero, as shown in Table 3.
Classifying Gamblers into Five Categories
Gamblers who had an indication of both Negative Consequences and Persistence
were placed into the Problem Gambler category. As a result of the path analysis,
three constructsImpaired Control Begin, Risky Practices Later and Preoccupation
Obsessionwere designated as Advanced Risk indicators based on two criteria:
Each was found to lead directly into one of the problem gambling constructs of
Negative Consequences or Persistence, and each was signicantly inuential on
either Negative Consequences or Persistence (Figure 1). Two constructs were
designated as Intermediate Risk indicators: Impaired Control Continue and Risky
Practices Earlier. Both led to Advanced Risk constructs and were linked in the
PLS-SEM model progressively to lower risk constructs.
Three constructs were used to identify Early Risk gamblers. Risky Cognitions
Beliefswasfoundonlytoinuence Risky Cognitions Motivesandwaspositioned
at the very start of the path with no other constructs inuencing it. Its overall effect
on either Negative Consequences or Persistence was lowest of all constructs, at 0.16
Table 2
Effect Size (f
) of Constructs on Selected Target Constructs
Risky Practices Later Negative Consequences Persistence
Risky Cognitions Beliefs 0.20 0.16 0.19
Risky Cognitions Motives 0.47 0.33 0.38
Preoccupation Desire 0.29 0.28 0.30
Risky Practices Earlier 0.29 0.14 0.28
Impaired Control Continue 0.32 0.34 0.32
Preoccupation Obsessed 0.32 0.25 0.36
Impaired Control Begin 0.34 0.54 0.34
Risky Practices Later 0.49 0.28
Negative Consequences 0.57
Note. Values of 0.35, 0.15 and 0.02 are considered to be large, medium and low effect levels.
and 0.19 respectively. As Risky Cognitions Motives is a formative construct and
located in the PLS-SEM model, where it affects both earlier and later constructs,
we used this construct to identify those gamblers who were earlier in the hierarchy
of risk for problem gambling. These individuals have yet to exhibit risky practices
or impaired control and yet had indications of risky cognitions. The intent was to
use this construct to identify persons who have a clear indication of risk before they
are gambling in a risky manner. Preoccupation Desire had fairly large levels of
inuence (0.280.30) on Negative Consequences and Persistence. As this construct
followed Impaired Control Motives and was found to inuence Impaired Control
Continue and Preoccupation Obsession, it was located early on the path to problem
Table 4 summarizes the criteria used for classifying machine gamblers to one of the
ve levels of risk for problem gambling.
Setting Criterion Levels for Constructs as Indicators
For all ten constructs analyzed using ROC analysis the statistical signicance for the
models was po0.000. The results are summarized in Table 5. Sensitivity ranged
from 41.4% to 90.5%. Preoccupation Obsession and Risky Cognitions Motives had
sensitivities of 41.1% and 43.1% respectively; Negative Consequences and Impaired
Control Continue had sensitivities of 90.5% and 83.1% respectively. Specicity
ranged from 81.1% to 99.0%, with Risky Cognitions Beliefs scoring the lowest and
Preoccupation Obsession scoring the highest.
Six of the constructs formed indicators based on cut offs of two: Persistence,
Preoccupation Obsession, Impaired Control Begin, Risky Practices Later, Risky
Cognitions Motives and Risky Cognitions Beliefs. The four constructs with cut offs
of three were Negative Consequences, Risky Practices Earlier, Impaired Control
Continue and Preoccupation Desire.
Table 3
Predictive Relevance of Latent Variables for Persistence
Predictive Relevance (Q
Risky Cognitions Beliefs 0.294
Risky Cognitions Motives 0.122
Preoccupation Desire 0.091
Risky Practices Earlier 0.246
Impaired Control Continue 0. 285
Preoccupation Obsessed 0.249
Impaired Control Begin 0.314
Risky Practices Later 0.315
Negative Consequences 0.247
A Prole of Indications for Gamblers at Each of the Five Risk Levels
Table 6 presents the percentage of respondents with indications of risk for the total
sample as well as for the ve FLAGS-EGM risk categories. For this particular
sample, the most prevalent risk indicator was Impaired Control Continue at 23.5%,
followed by three indicators with a similar prevalence: Preoccupation Desire
(18.4%), Risky Practices Earlier (17.9%) and Risky Cognitions Motives (17.1%).
Risky Practices Later (12.8%), Risky Cognitions Beliefs (8.8%) and Impaired
Control Begin (8.6%) had a relatively low prevalence in this sample. Only 3.7% of the
sample had an indication of Preoccupation Obsession.
Problem Gamblers. As shown in Table 6, individuals classied as Problem
Gamblers because of indications of both Negative Consequences and Persistence
Table 4
FLAGS Five Levels of Player Risk for Machine Gambling
Label Description
Level V Problem
A Problem Gambler is a person who agged as exhibiting both Negative
Consequences and Persistence and is characterized as having experienced
harm in association with gambling yet is persisting in gambling.
Level IV Advanced
Those persons at Advanced Risk are not agging as a Problem Gamblers
(i.e., scoring on Negative Consequence and Persistence) but hold one or
more indications on the ve constructs directly connected to either
Negative Consequences or Persistence. Three of these constructs are
Impaired Control Begin, Preoccupation Obsessed and Risky Practices
Later. Negative Consequences and Persistence are included as it is
possible that a person only agged on one of these constructs and,
therefore, has not (yet) reached the threshold for identication as a
problem gambler.
Level III Intermediate
Those at Intermediate Risk are not Problem or Advanced Risk gamblers,
but have been agged on one or more of the Intermediate Risk constructs.
The Intermediate Risk constructs are Impaired Control Continue and
Risky Practices Earlier. Intermediate Risk Gamblers are not triggering on
Negative Consequences or exhibiting signs of Persistence. While higher in
the risk hierarchy than the Early Risk Gamblers these players comprise
individuals at pre-harm risk levels.
Level II Early Risk Those at Early Risk have agged on at least one of Risky Cognitions
Beliefs, Risky Cognitions Motives or Preoccupation Desire but are not
triggering the Advanced Risk or Problem Gambling constructs and are
also characterized as a pre-harm risk group.
Level I No Indication
of Risk
Those at No Indication of Risk do not ag on any of the risk indicators
although it is possible that they answered yes to one or more statements
making up some of the constructs. For those subjects who answered yes
to at least one statement there was insufcient certainty for us to say
there was an indication on one of the dimensions.
Level 0 Non-Gambler A Non-Gambler is at no-risk currently because he or she is not now
engaging in behaviours that could lead to harm.
all had an indication of Impaired ControlContinue(100%).Theywerealsolikely
to have indications of Risky Practices Earlier (89.7%), Risky Practices Later
(82.8%) and Risky Cognitions Motives (82.8%). A large proportion of the
Problem Gamblers also had indications of Impaired Control Begin (65.5%) and
Preoccupation Desire (62.1%). Preoccupation Obsession (41.1%) and Risky
Cognitions Beliefs (31.0%) were common, but not frequent indicators for
Problem Gamblers.
Advanced Risk. Some of those categorized as Advanced Risk had indications of
Persistence (33.3%) or Negative Consequences (22.1%), but not both, as this would
have categorized them as Problem Gamblers. The most common indicators of risk
for this category were Impaired Control Continue (69.4%), Risky Practices Later
(66.7%), Risky Practices Earlier (55.6%) and Preoccupation Desire (55.6%). Both
Risky Cognitions Motives (38.9%) and Impaired Control Begin (36.1%) were fairly
prevalent among the Advanced Risk Gamblers, while Risky Cognitions Beliefs
(22.2%) and Preoccupation Obsession (14.3%) were less prevalent.
Intermediate Risk. Impaired Control Continue was the most prevalent indicator
(77.3%) for those in the Intermediate Risk category. Also, somewhat important were
Risky Practices Earlier (47.7%) and Preoccupation Desire (45.5%). Both Risky
Table 5
Results of ROC Analysis for Ten FLAGS-EGM Constructs
Construct State Variable Value Cut Off
Sensitivity Specicity Area
Persistence PGSI 8+2 11.0% 76.2% 97.2% 95.0%
PGSI 8+3 9.9% 90.5% 94.9% 95.6%
FLAGS-EGM PG 2 3.7% 41.4% 99.04% 78.1%
Impaired Control
FLAGS-EGM PG 2 8.6% 65.5% 96.2% 91.4%
Risky Practices
FLAGS-EGM PG 2 12.8% 82.8% 93.0% 95.0%
Risky Practices
Advanced Risk
3 17.9% 70.8% 93.2% 89.0%
Impaired Control
Advanced Risk
3 23.5% 83.1% 89.0% 89.8%
Intermediate Risk
3 18.4% 53.2% 95.8% 82.7%
Risky Cognitions
Intermediate Risk
2 17.1% 43.1% 93.6% 75.7%
Risky Cognitions:
FLAGS Intermediate
2 27.3% 47.7% 81.1% 68.0%
Cognitions Motives (20.5%) and Risky Cognitions Beliefs (4.5%) have low
prevalence among those in this category.
Early Risk. For Early Risk Gamblers, the key indicators were Risky Cognitions
Motives (43.6%), followed by Risky Cognitions Beliefs (35.9%) and Preoccupation
Desire (28.2%).
Comparison of FLAGS-EGM to the PGSI
The overall distribution by risk categories was somewhat similar for the two
measures (Table 7). FLAGS-EGM identied 60.4% as No Detectable Risk, 10.4% as
Early Risk, 11.8% as Intermediate Risk, 9.6% as Advanced Risk and 7.8% as
Problem Gamblers. The PGSI identied 54.8% as No Risk, 19.3% as Low Risk,
20.3% as Medium Risk and 5.6% as Problem Gamblers. The PGSI found 39.6% of
the sample to be at risk compared to 31.8% for the FLAGS-EGM.
Comparison of the classication Problem Gambler by the two instruments produced
a tetrachoric correlation of 0.947, indicating a very high degree of agreement
between the two instruments in terms of identifying problem gamblers. Using the
Table 6
Percent of FLAGS-EGM Risk/PG Segments with Specic Indications of Risk
Constructs All Gamblers
N = 374
No Risk
N = 226
Early Risk
Risk N = 44
Risk N = 36
Persistence 11.0% 0.0% 0.0% 0.0% 33.3% 100%
9.9% 0.0% 0.0% 0.0% 22.2% 100%
3.7% 0.0% 0.0% 0.0% 14.3% 41.4%
Control Begin
8.6% 0.0% 0.0% 0.0% 36.1% 65.5%
Risky Practices
12.8% 0.0% 0.0% 0.0% 66.7% 82.8%
23.5% 0.0% 0.0% 77.3% 69.4% 100.0%
Risky Practices
17.9% 0.0% 0.0% 47.7% 55.6% 89.7%
18.4% 0.0% 28.2% 45.5% 55.6% 62.1%
17.1% 0.0% 43.6% 20.5% 38.9% 82.8%
8.8% 0.0% 35.9% 4.5% 22.2% 31.0%
PGSI as the ‘‘ gold standard’’ for categorizing an individual as a Problem
Gambler, the sensitivity of the FLAGS-EGM measure was 85.7%, while the
specicity was 96.9%. The PGSI identied 21 problem gamblers in the sample
of 374 while the FLAGS-EGM identied 29. Eleven (37.9%) of those identied
as problem gamblers by the FLAGS-EGM were not categorized as such by
the PGSI. Only three individuals were categorized as problem gamblers by
the PGSI and not by the FLAGS-EGM. This particular result produced an
overlap of only 56.2% subjects being identied as a Problem Gambler by both
The overlap of those gamblers categorized at any level of risk by either instrument
(with the FLAGS-EGM Intermediate Risk and Advanced Risk categories
combined) was 33.9%. The greatest source of discrepancy between the two measures
occurred when a gambler was categorized as At Risk by one instrument and at No
Risk or No Detectable Risk by the other. This particular inconsistency happened in
21.1% of the cases.
Table 8 presents the prole of the four discrepancy segments in terms of the
percentage of gamblers in those segments having indications on each of the ten
FLAGS-EGM constructs, as well as the percentage of those gamblers responding
either sometimes or more often to each of the nine PGSI statements.
PLS-SEM was used successfully to create a model to identify risk for problem
gambling and to classify an individual as a Problem Gambler. The model utilized all
ten constructs developed for this purpose and passed the four tests specied by
Table 7
Overlap in Classication by Risk Categories Between FLAGS-EGM and PGSI
No Detectable
No Risk 47.1% 6.4% 0.8% 0.5% 0.0% 54.8%
176 24 3 2 0 205
Low Risk 11.2% 2.4% 4.3% 1.1% 0.3% 19.3%
42 9 16 4 1 72
Medium Risk 2.1% 1.6% 6.7% 7.2% 2.7% 20.3%
8 6 25 27 10 76
0.0% 0.0% 0.0% 0.8% 4.8% 5.6%
0 0 0 3 18 21
Total 60.4% 10.4% 11.8% 9.6% 7.8% 100.0%
226 39 44 36 29 374
Table 8
Comparisons of Discrepancy Segments
EGM Higher
FLAGS (% agged on construct) (n= 50) (n= 29) (n= 21) (n= 10)
Risky Cognitions Beliefs 0.0% 44.8% 0.0% 10.0%
Risky Cognitions Motives 0.0% 37.9% 33.3% 80.0%
Preoccupation Desire 0.0% 24.1% 42.9% 30.0%
Risky Behaviours Earlier 0.0% 10.3% 42.9% 80.0%
Impaired Control Continue 0.0% 17.2% 57.1% 100.0%
Impaired Control Begin 0.0% 3.4% 0.0% 40.0%
Risky Behaviours Later 0.0% 3.4% 23.8% 60.0%
Preoccupation Obsession 0.0% 0.0% 0.0% 20.0%
Negative Consequences 0.0% 6.9% 0.0% 100.0%
Persistence 0.0% 6.9% 4.8% 100.0%
PGSI (% responded sometimes or more
You bet more on the slot machines at a
casino than you could really afford to
54.0% 0.0% 42.9% 90.0%
You needed to gamble on the slot
machines at a casino with larger
amounts of money to get the same
feeling of excitement?
14.0% 0.0% 19.0% 30.0%
When you gambled on the slot machines
at a casino, you went back another day
to try and win back the money you lost?
38.0% 0.0% 33.3% 60.0%
You borrowed money or sold anything to
get money to gamble on the slot
machines at a casino?
2.0% 0.0% 4.8% 30.0%
You felt that you might have a problem
with gambling on the slot machines at a
10.0% 0.0% 23.8% 90.0%
People have criticized your betting or told
you that you had a gambling problem
with slot machines at a casino,
regardless of whether or not you
thought it was true?
14.0% 0.0% 14.3% 50.0%
You have felt guilty about the way you
gamble, or what happens when you
gamble on the slot machines at a casino?
24.0% 0.0% 14.3% 80.0%
Your gambling on slot machines at a
casino has caused you any health
problems, including stress or anxiety?
2.0% 0.0% 0.0% 10.0%
Your gambling on slot machines at a
casino has caused any nancial
problems for you or your household?
0.0% 0.0% 0.0% 40.0%
Urbach and Ahlemann (2010). Specically, the coefcients of determination were
sufcient, the path coefcients were signicant, the independent latent variables had
medium impact on dependent latent variables, and all constructs had a predictive
relevance greater than 0.0. In addition, ROC analyses identied the optimal cut off
to form indicators of risk that met our criteria. As a result of these analyses, gamblers
were classied as being in one of ve categories as described in detail immediately
Problem Gamblers
Indications of both Negative Consequences and Persistence were required before an
individual was considered to be a Problem Gambler. It should be noted that in
creating the Negative Consequence construct, we did not address all possible forms
of harm. We did not ask questions about aggressive or illegal behaviours,
relationship problems, mental illness or attempted suicide. We hypothesized that
individuals suffering these more severe consequences of gambling would also ‘‘ ag’’
on the less severe and more generally-phrased statements. Moreover, queries of this
nature could be viewed as threatening and therefore left unanswered by some
respondents or lead others to stop participating in the survey altogether. Researchers
who want to know the prevalence of these more severe consequences could include
additional questions but would need to exclude them when deriving the Negative
Consequences indicator.
The FLAGS-EGM instrument categorized 7.8% of the individuals as problem
gamblers compared with 5.6% as identied by the PGSI for the same sample.
Without further research, it cannot be determined if the FLAGS-EGM would always
have a higher identication rate. It may be that the fourteen negative-consequence
statements triggered recognition on the part of respondents as to the harms they had
experienced, thereby identifying gamblers who would not in fact be classied as
problem gamblers using the PGSI.
Advanced Risk
Individuals with Advanced Risk could have indications of either Persistence or
Negative Consequences in contrast with Problem Gamblers who displayed both
characteristics. Preoccupation Obsession, Risky Practices Later and Impaired
Control Begin were also associated with Advanced Risk (Table 6). A third of the
Advanced Risk group also had an indication of Persistence. This nding is an
important indicator of risk in this group as it identies those respondents who admit
they intend to continue gambling despite the fact that it will lead to further harms.
Although 22.2% of those designated as being at Advanced Risk had an indication of
Negative Consequences, they were not in fact persisting in gambling. There are
several possible explanations for this. Some gamblers, once they had experienced
harms, may in turn have found ways to control their gambling behaviour. Others
may have stopped gambling within the one-year time frame designated in the
instrument and thus did not have an indication of persistence. Still others may have
only recently experienced harms and had yet to become persistent in their gambling
behaviour. Regardless, however, of their respective particular situations, these
gamblers were nonetheless placed in the Advanced
Risk category because they experienced Negative Consequences. The mere
possibility of their relapsing or persisting in gambling was sufcient to warrant
assigning them to this category.
Impaired Control Begin inuenced Risky Practices Later and also had a direct and
fairly strong impact on Negative Consequences (Fig. 1). The Responsible Gaming
Device (RGD) and RG Tracking System, designed by Techlink Entertainment
Systems and tested in Windsor, Nova Scotia (Schellinck & Schrans, 20067), includes
built-in self-exclusion features, as well as an option to track expenditures over
extended periods; this specic option could help a gambler overcome Impaired
Control Begin. This feature would also be particularly useful wherevia wide-area
networks in smaller venues, such as bars and clubsEGMs are provided. Assistance
delivered by Gamblers Anonymous and counsellors, as well as venue exclusion
programs, are specically designed to work with individuals with this degree of
impaired control.
Risky Practices Later also had a strong connection to Negative Consequences
(Fig. 1). By limiting the amount of money that can be borrowed on the premises, and
by ensuring that loan sharks are kept away, the operators or staff of venues could be
of help to gamblers who are attempting to borrow money. Pre-commitment could be
effective in assisting such individuals reduce these highly risky practices.
Intermediate Risk
The reective construct Impaired Control Continue and the formative construct
Risky Practices Earlier were the indicators of Intermediate Risk (Table 6). Impaired
Control Continue, in turn, had a strong inuence on Risky Practices Earlier and
Impaired Control Begin (Fig. 1). This latter relationship suggests that gamblers rst
lose control during a session, and then later lose control between sessions because of
an inability to resist gambling again.
Both Impaired Control Continue and Risky Practices Earlier were mainly associated
with behaviours that would occur ‘‘ on the oor,’’ and which gamblers themselves
could potentially cut back on. Consequently, interacting with individuals on location
during a gambling session could be important in reducing their risk levels. These
gamblers would most likely benet from responsible gambling features such as the
Live Action component of the My Play system (Schellinck & Schrans, 2007;
Schellinck & Schrans, 2011). This feature provides a gambler with real-time
monitoring of gambling activities, including cumulative spending during the session.
Such a nding provides support for a role of My Play or a similar program to help
individuals control their risky behaviours once identied.
Gamblers who were administered the FLAGS-EGM instrument have been shown to
be motivated to control their gambling (Buckley, 2013). Providing gamblers access to
the FLAGS-EGM while on the oor, either in the form of a pamphlet (Buckley,
2013) or on screen through the EGM interface, could be helpful in reducing risky
practices in the casino.
Early Risk
A respondent needed to endorse statements associated with Risky Cognitions
Motives, Risky Cognitions Beliefs or Preoccupation Desire to be classied as an
Early Risk gambler. A relatively large proportion of the entire sample, i.e., 17.1%,
had, regardless of their level of risk, an indication of Risky Cognitions Motives. In
contrast, Risky Cognitions Beliefs appeared to be a risk factor early for certain
gamblers (Fig. 1). Risky Cognitions Motives, with the largest number of paths
leading from it in the PLS-SEM model inuenced Preoccupation Desire,
Preoccupation Obsession, Risky Behaviour Earlier and Risky Behaviour Later,
and was the third most common indicator for the Problem Gamblers. These results
suggest that it may be more effective to lower an individuals risk by reducing risky
motives than by reducing risky beliefs. Moreover, these results emphasize the need
for investigation into the factors underlying motives in the effort to decrease risk due
to gambling.
Preoccupation Desire, experienced by 62.1% of the Problem Gamblers, had a very
strong inuence on Impaired Control Continue, the most common indicator for risk.
Reducing the entertainment value of the gambling experience, changing the
reinforcement schedule by changing the frequency and nature of wins, or reducing
the marketing material received by the gambler, might each in turn affect the
gamblers desire to gamble. Gamblers have little control over these inuences, and
changes to these elements that might reduce Preoccupation Desire will likely need to
be initiated by the gambling providers and regulators.
No Detectable Risk
For gamblers to be placed in the No Detectable Risk category, FLAGS-EGM would
not have given any indications of risk or harms because of gambling within the last
year. Gamblers may have endorsed a few of the statements but not enough of them
within a construct to provide sufcient evidence for an indication of risk on that
criterion. Certain of these gamblers may indeed have been at risk or problem
gamblers in the past but at least for the previous year they are not in these categories.
Certain of these gamblers will also have personal or situational factors that could
lead to higher risk levels associated with gambling, levels that we have not measured.
Nonetheless, these factors have as yet not manifested themselves in terms of
cognitions about gambling, impaired control, preoccupation, risky practices or
harms and persistence, and therefore are likely to remain at that risk level until
something in their environment triggers changes that lead to elevated risk levels. This
instrument would at that time identify these persons.
Comparison of the FLAGS-EGM to the PGSI
Overall, the sizes of the risk segments created by FLAGS-EGM and the PGSI are
similar. However, the number of Problem Gamblers identied by the FLAGS-EGM is
somewhat larger. The two instruments differed markedly in terms of assigning individuals
to specic levels of risk. In fact, they agreed in only about one-third of the cases. As well,
one-fth of those in the sample were assigned to a risk category by just one of the
instruments. The prole of four discrepancy segments (Table 8) provides more insight as
to why the two instruments classied gamblers into different risk categories. As discussed
below, these results suggest FLAGS-EGM is a more appropriate instrument for
identifying and categorizing gamblers at risk of becoming a Problem Gambler.
Segment One. Segment One comprised those individuals who were categorized
as at risk (Low Risk and Medium Risk) by the PGSI and No Detectable Risk by the
FLAGS-EGM. These persons may rarely exhibit the characteristics identied in the
PGSI as 94% of the responses by those in this segment were ‘‘ Sometimes.’’ The single
largest contributor (54.2%) to designating these persons as at risk by the PGSI is the
statement ‘‘ you bet more than you could really afford to lose.’’ In a casino
environment overspending may occur ‘‘ sometimes’’ for a variety of reasons unrelated
to risk. For example, friends are not ready to leave, the bus is not ready to go, a
friend is winning or the environment is exciting. FLAGS-EGM has a similar
statement but it does not categorize gamblers, based on this statement alone, as at
risk. The PGSI also classies gamblers as at risk if they ‘‘ sometimes’’ feel guilty
about their gambling; a number of gamblers (24%) in this segment indicated that
they sometimes feel as this. These feelings of guilt may be mediated by just being
present in the casino environment regardless of ones gambling behaviour.
Individuals are cautioned constantly to ‘‘ play within their limits,’’ or ‘‘ gamble
responsibly.’’ This reminder could lead them to agree that they sometimes feel guilty
about their gambling.
The statement in the PGSI ‘‘ went back another day to try and win back the money
you had lost’’ is meant to identify chasing behaviour. However, as worded, it may
have led individuals to endorse this behaviour and thus erroneously led to the PGSI
classifying gamblers as Low Risk based on behaviour that is not in fact really chasing
behaviour. For example, gamblers who (1) visit the casino more frequently and thus
gamble more often in consecutive days, or (2) regularly gamble over two days of the
weekend, or (3) go to a casino as a destination, and therefore play for several days in
a row, are, in all three cases, more likely to answer ‘‘ sometimes’’ to this statement.
When asked about their intentions when returning to gamble, such gamblers may
admit to wanting ‘‘ sometimes’’ to win back the money they had previous lost even
though the behaviour was not problematic. To reduce such a frequency bias, the
equivalent FLAGS-EGM statement included the phrase ‘‘ after losing more money
than I wanted.’’ The FLAGS-EGM also addressed and claried the more specic
aspects of the situation (‘‘ I usually try to win it back by playing again either later that
day or on another day’’ ). In general, gamblers classied as Low Risk by the PGSI
but at No Detectable Risk by FLAGS-EGM have answered ‘‘ sometimes’’ to
questions that may have relatively low thresholds for casino gamblers. Using such a
statement could inappropriately classify someone as having a risk for problem
gambling based on such a criterion.
Additionally, individuals categorized as Low Risk by the PGSI but as No Detectable
Risk by FLAGS-EGM were only required to answer ‘‘ sometimes’’ to one question on
the PGSI to be identied as Low Risk. These individuals made up approximately two-
thirds (63.9%) of PGSI Low Risk Gamblers in this sample and nearly a third (31.1%) of
all of those identied by the PGSI as at risk. The PGSI and other instruments that rely
on a continuum based on a sum score starting at 1 are effectively categorizing persons
based on the endorsement of a single statement. In contrast, individuals that endorsed no
more than one statement in all the constructs in FLAGS-EGM were assigned to the No
Detectable Risk category. In line with the arguments for using multi-item measures to
identify latent constructs (Churchill, 1979) a minimum score of 2i.e., endorsement of
two questions or morewasneededtosaywith condence that a person holds an
indication of risk on any one of the ten constructs in the FLAGS-EGM.
Segment Two. This group comprised those gamblers not designated by the
PGSI as at risk but designated by FLAGS-EGM as Early Risk. Those gamblers who
agged on any of three indicators, Risky Cognitions Beliefs, Risky Cognitions
Motives and Preoccupation Desire (Table 8), but did not ag on the more advanced
indicators of risk, were designated as Early Risk gamblers. The PGSI does not have
statements that cover these indicators. However, the development of the FLAGS-
EGM constructs was based on an extensive review of the literature as well as
previous research with samples of the gambling population that identied these
indicators as risk factors to test and our PLS-SEM analysis showed them to be
signicantly related to the development of problem gambling. This nding suggests
that administering this instrument could provide valuable predictors of individuals at
risk for problem gambling that are not currently being assessed by the PGSI.
Segment Three. These gamblers were classied as Low Risk by the PGSI but
are classied as Intermediate or Advanced Risk gamblers by the FLAGS-EGM. The
concerns and potential bias associated with the statements in the PGSI, and the
alternative approach used in the FLAGS-EGM, can be found in the discussion on
Segment One and will not be further described here.
In this case, because most of the responses to the PGSI statements are ‘‘ sometimes,’’
the total PGSI score does not exceed 2 for these gamblers, placing them in the Low
Risk category. In FLAGS-EGM, sometimes exhibiting higher-risk characteristics is
deemed to be sufcient reason to place them in a higher risk category. The point is
this. By not treating all statements equally in terms of risk indicationi.e., the words
‘‘ sometimes,’’ ‘‘ often’’ and ‘‘ frequently’’ are used in the statements to weight
effectively the statement itself, and the statements are placed in constructs that are
associated with different risk levelstheFLAGS-EGMcanbetterassignapersontoa
risk category, based on both the extent and the riskiness of the behaviour. This feature
means that the gambler, to have an indication of Intermediate Risk, often needs to spend
more time gambling than intended but only needs sometimes to borrow money from
other gamblers to be classied as an indication of Advanced Risk.
Segment Four. Gamblers who were classied as Medium Risk by the PGSI and
Problem Gamblers by the FLAGS-EGM were included in this category. Almost all
of these gamblers (90%) indicated through the PGSI that they believed they had a
gambling problem ‘‘ sometimes’’ but this was not enough to categorize them as
Problem Gamblers. The difference is because all participants identied as Problem
Gamblers in the FLAGS-EGM indicated experiencing at least three harms caused by
gambling and exhibited Persistence. In the PGSI, if we assume the last four
statements all have to do with negative consequences then the respondents could say
they sometimes experience these consequences, but this is not sufcient to move them
into the problem gambling category. In the FLAGS-EGM for example, having had
problems paying off debts in the last year at all, or sometimes having to juggle money
and bills in order to gamble, are indications of negative consequences that led them
to be classied as Problem Gamblers. These gamblers indicated experiencing three or
more of the fourteen consequences listed in FLAGS-EGM and therefore met the
Negative Consequences criterion for Problem Gambler. The main consequences
indicated were: (1) They do not want others to know about their gambling
behaviours; (2) they feel depressed about their gambling; (3) they believe gambling
has interfered with their lifes goals, and (4) they do not like the type of person they
have become. This nding suggests that the PGSI may not be identifying negative
consequences that are related to the gamblers self-perception and state of mind.
A key difference between the FLAGS-EGM classication scheme and that of other
instruments such as the PGSI and SOGS is that with the FLAGS-EGM the gambler
is classied based on the nature of the indicators agged. With the former screens,
the gambler is classied based on a summed score. Thus, individuals who indicated
that they sometimes borrow money to gamble could be placed at a low risk level by
the PGSI (depending upon what other statements they endorsed). In comparison, the
FLAGS-EGM would consider borrowing money a high-risk behaviour that
consequently identied a gambler as Advanced Risk. We believe this method of
classication is a major strength of the FLAGS-EGM instrument.
Our analysis provides strong evidence that people will progress towards problem
gambling if they have these indicators as described in this study. To test this
hypothesis further, we need a longitudinal study to measure the movement of
gamblers between risk groups over time. Of course, this research is based on a
convenience sample of Ontario slot players, and further research is consequently
needed to determine if the relationships identied here exist in other jurisdictions.
Furthermore, our research applies only to EGM players. When developed in
Schellinck, T. et al. (in press), the Preoccupation Obsession construct passed all
validity and reliability tests, except for Composite Reliability, which requires at least
three items in the construct to produce an accurate statistic. We therefore used the
two statement version of the construct in this phase of the analysis in order to
construct the FLAGS-EGM instrument. Further research is being conducted with a
revised version of the construct containing four items.
It is often suggested that when conducting the ROC analysis the state variable should
be a ‘‘ gold standard.’’ When conducting ROC analysis on the Negative Consequences
and Persistence constructs we used the PGSI score of 8+as our state variable.
However, we are not aware of any validated risk measures that could be used as gold
standards, and therefore, as state variables for the constructs in each preceding risk
level, had to rely on the risk level indicators already created using ROC analysis within
FLAGS-EGM. Thus, those gamblers already classied as Intermediate Risk were the
state used in the ROC analysis when analysing the three Early Risk constructs, and the
sample size of 309 was more than sufcient for this purpose.
The instrument was designed to identify at risk and Problem Gamblers who are at
risk due to EGM gambling. As such, it was done to make the terms in the statements
more exact and therefore the instrument more accurate and easier to self-administer.
This fact means it cannot be used to measure risk due to other forms of gambling.
However, in many jurisdictions that offer wide-area network gambling, EGM
gambling is the primary form of gambling, and thus needs to be studied separately.
Our experience is that regulators, whose jurisdiction includes a large number of
EGMs, as well as gambling providers such as casinos and betting shops, want to
measure carefully the impact of EGMs on risk and Problem Gambling exclusive of
table games, sports betting and lotteries, and as such the FLAGS-EGM will nd
many applications.
Using SEM-PLS analysis, we have created an instrument that should provide reliable
information to EGM gamblers concerning their risk levels. As summarized below,
the FLAGS-EGM has several key characteristics that make it very suitable for use as
a measure of gambling risk and harm.
First, based upon its design, the instrument should be highly effective in identifying
individuals at risk due to EGM gambling. We have chosen indicators that are proven
to be associated with problem gambling (beliefs, motives, impaired control,
preoccupation, consequences and persistence). Moreover, the individual must be
an active gambler to respond to the statements.
Second, the FLAGS-EGM is easily administered, either by gamblers, themselves, or
in a clinical context. Gamblers understood the statements, interpreted them
consistently and believed that they were relevant to their situation (Schellinck,
T. et al., in press). Consequently, both gamblers and health providers should be able
to assess in an accurate and informative manner individual risk levels caused by
Third, the instrument could be set up as a responsible gambling (RG) module on
gambling machines or players could be invited to ll out the FLAGS-EGM on the
Internet at an RG site. When administered via computer the number and nature of
risk indicators, and the level of their risk associated with gambling, could be
provided automatically to the gambler. Fourth, and perhaps most important,
the instrument could provide policy makers with detailed information as to the
nature of risk faced by gamblers. Using the FLAGS-EGM in this manner could lead
to effective solutions for reducing the potential for the harms associated with
Bliemel, M., & Hassanein K. (2007). Consumer satisfaction with online health
information retrieval: A model and an empirical study. e-Service Journal, 5(2),
Bonett, D. G., & Price, R. M. (2005). Inferential Methods for the Tetrachoric
Correlation Coefcient. Journal of Educational and Behavioral Statistics, 30(2),
213225. doi: 10.3102/10769986030002213
Buckley, M. F. (2013). A self-administered problem gambling screen (FLAGS II) as
motivational intervention for problem gamblers (Doctoral dissertation). Available
from ProQuest Dissertations & Theses database (UMI No. 3601136).
Chin, W. W. (1998). Issues and opinion on structural equation modeling. MIS
Quarterly, 22(1), vii-xvi.
Chin, W. W. & Newsted, P. R. (1999). Structural equation modeling analysis with
small samples using partial least squares. In R. H. Hoyle (Ed.), Statistical strategies
for small sample research. Thousand Oaks, CA: Sage (pp. 307341).
Churchill, G. A., Jr. (1979). A paradigm for developing better measures of marketing
constructs. Journal of Marketing Research, 16(1), 6473.
Cohen, J. C. (1992). A power primer. Psychological Bulletin, 112(1), 155159.
Conigrave, K. M., Hall, W., & Saunders, J. B. (1995). The AUDIT questionnaire:
Choosing a cut-off score. Addiction, 90(10), 13491356. doi: 10.1111/j.1360-
Ferris J., & Wynne, H. (2001). The Canadian problem gambling index: Final report.
Ottawa, ON: Canadian Centre on Substance Abuse.
Hair, J. F., Hult, T. M., Ringle, C. M., Sarstedt, M. (2013). A primer on partial least
squares (PIs) path modeling. Thousand Oaks, CA: Sage.
Hair, J. F., Ringle, C. M., & Sarstedt, M. (2011). PLS-SEM: Indeed a Silver Bullet.
The Journal of Marketing Theory and Practice, 19(2), 139152.
Maddern, C., & Rogala, M. (2006). Testing of the self-administered problem
gambling scale prototype screen: A draft report. Melbourne, AU: Market Solutions
PTY. Retrieved March 25, 2015, from
Metz, C. E. (2006). Receiver operating characteristic analysis: A tool for the
quantitative evaluation of observer performance and imaging systems. Journal of the
American College of Radiology, 3(6), 413422. doi: 10.1016/j.jacr.2006.02.021
Ringle, C. M., Wende, S., & Will, A. (2014). SmartPLS. Hamburg, DE: University
of Hamburg. Retrieved from
Schellinck, T. (2006). Phase II: Quantitative analysis for the Victoria self-administered
problem gambling screen. Victoria, AU: Victoria Department of Justice. Retrieved
March 25, 2015, from
Schellinck, T. & Schrans, T. (2007). Assessment of the behavioural impact of
responsible gaming device features: Analysis of Nova Scotia player-card data: Windsor
trial. Nova
Scotia Gaming Corporation, Halifax, NS. Retrieved March 25, 2015, from http://
Schellinck, T., & Schrans, T. (2010). Evaluating the impact of the ‘‘ My-Play’’
system in Nova Scotia: Phase I: Regular VL benchmark survey: Technical report.
Halifax, NS: Nova Scotia Gaming Foundation. Retrieved March 25, 2015,
Schellinck, T., Schrans, T., Schellinck, H., & Bliemel, M. (in press). Construct
development for the FocaL Adult Gambling Screen for Electronic Gambling
Machine players (FLAGS-EGM): A measurement instrument for risk due to
gambling harm and problem gambling associated with electronic gambling
machines. Journal of Gambling Issues.
Uebersax, J. S. (1987). Diversity of decision-making models and the measurement of
interrater agreement. Psychological Bulletin, 101(1), 140146.
Urbach, N., & Ahlemann, F. (2010). Structural equation modeling in information
systems research using partial least squares. Journal of Information Technology
Theory and Application, 11(2), 540.
Risky Cognitions Beliefs
You can sometimes tell when the machine is about to pay out big because the
symbols start getting closer to lining up on the pay line (e.g., almost winning).
I feel the machines are xed sometimes so that you cant win on them.
It is important for me to use a system or a strategy when I play the machines.
I believe that in the long run I can win playing slots at the casino.
If a slot machine hasnt had a big pay out in a long time, it is more likely to do so
Risky Cognitions Motives
I sometimes play the slots in hopes of paying off my debts/bills.
I sometimes play the slots when Im feeling down or depressed.
Gambling on the slots is a way I can try to get some money when I need it.
can escape by playing the slots whenever I am worried or under stress.
Preoccupation Desire
If I could play the machines all the time I would.
I wish I could gamble on the slots more often.
I would like to play the slots almost every day.
I like to play the slot machines every chance I get.
Preoccupation Obsession
I sometimes dream about playing the slot machines.
I spend more time than I used to thinking about playing the slots.
Risky Practices Earlier
I sometimes exceed the amount of money I intended to spend in order to win back
money I have lost.
When gambling on the slots I usually use my bank or debit card to get more money
so I can keep playing.
I play max bet if Im on a winning streak.
If I win big I am likely to put the money back into a machine and keep playing.
When gambling on a slot machine I usually play as fast as I can.
I have sometimes gambled for more than six hours straight when I was playing the
Risky Practices Later
After losing more money than I wanted on the slots I usually try to win it back by
playing again either later that day or on another day.
When gambling on the slots I usually use my credit card to get more money so I
can keep playing.
When I gamble with friends or family I sometimes stay and continue to play after
they have stopped or left.
I have sometimes borrowed money so I could go and gamble on the slots.
I have borrowed money from other people at the casino in order to continue
I have left the casino to get more money so I can come back and keep on gambling.
Impaired Control Continue
I often spend more money gambling than I intended.
Even when I intend to spend a few dollars gambling, I often end up spending much
I sometimes gamble with money that I cant really afford to lose.
Once I have started gambling on the slots I nd it very hard to stop.
I often spend more time gambling than I intend to.
Impaired Control Begin
I have tried to cut back on my slots play with little success.
I have tried unsuccessfully to stop or reduce my gambling on the slots.
There have been times I have gambled despite my desire not to.
Negative Consequences
My goals in life have been jeopardized by my slot play.
I often cant sleep because I am worrying about my slot machine gambling.
I have had problems paying off debts accumulated from playing the slots.
Since I started playing the slots I dont like the type of person I have become.
Sometimes I have to juggle money and bills to cover the cost of my slot machine
I wouldnt want anyone to know how much time or money I spend at the casino.
Sometimes I feel depressed over my slots play.
Others are disappointed in me because of my gambling.
I have friends or family who are concerned about my slots play.
I have sometimes missed events or neglected family, friends or work in order to
play the slots.
When I leave the casino, I have sometimes been short of cash for parking, food, or
a ride home.
I have become somewhat of a loner because of my slot gambling.
I sometimes have spent time gambling on the slots when I was supposed to be
doing something else important.
My gambling has caused me to have a falling out with the people I used to hang
out with.
I continue to play the machines despite experiencing problems or other negative
I continue to gamble despite the bad things that happen to me.
I gamble even though I know it is likely to lead to problems for me.
Even if money is tight, I continue to play the slots to get big wins.
Manuscript history: Submitted October 16, 2012; Accepted October 27, 2014.
For correspondence: Tony Schellinck, PhD, Focal Research Consultants Limited, 7071
Bayers Rd., Suite 326, Halifax, NS B3L 2C2, E-mail:,
Website address:
Competing interests: None declared.
Ethics approval: The Ontario Institutional Review Board (ON IRB). Final protocol
approval was obtained for ‘‘ Preliminary Development of a Self Administered
Gambling Risk Assessment Instrument for Slots’’ on June 23 2008.
Funding: Ontario Problem Gambling Research Centre: Grant # 2755.
Contributors: T. Schellinck planned the document. T. Schellinck and HS drafted and
wrote the manuscript with editorial contributions from T. Schrans and MB. HS and
T. Schrans conducted the gambling-related literature review. T. Schrans conceptua-
lized the research design and conducted the focus group and survey studies. T.
Schellinck and MB assessed the current analytical literature and designed the
analysis approach. T. Schellinck conducted the analysis and nalized the design of
the constructs.
Dr. Tony Schellinck is an Adjunct Professor in the Faculty of Graduate Studies and
the Rowe School of Business at Dalhousie University, Canada, as well as CEO of
Focal Research Consultants Limited. From 1996 to 2013 he was the F. C. Manning
Chair in Economics and Business at Dalhousie University. Since 1989 he has
conducted research into gambling behaviour for industry, government, public health
and regulatory agencies. This work included a ten-year large-scale monthly tracking
study of gambling behaviour, over 300 focus group sessions with gamblers, the 1998
Nova Scotia Video Lottery Study, two large scale studies into the value of
responsible gambling features on VLT machines, and the Nova Scotia Adolescent
Gambling Exploratory Research: Identication of Risk and Gambling Harms
Among Youth. Dr. Schellinck worked on creating the rst algorithms deployed in
casinos that identied using player loyalty data high-risk gamblers.
Ms. Tracy Schrans is Principal and President of Focal Research Consultants an
independent research rm in Halifax, NS. Over the last twenty years Tracy has
conducted numerous government, public health, and industry-sponsored research
projects on a wide range of issues, with a particular emphasis on gambling- and
alcohol-related issues. She consults internationally in responsible gambling and
corporate social responsibility, social policy, player tracking and loyalty data
analysis. Ms. Schrans is one of the developers of new instruments for measuring pre-
harm risk for gambling among adults (FLAGS-EGM and FLAGS General) and
adolescents (FYGRS) for prevention applications. She continues to work at the
forefront of gambling behavior analytics, assisting gambling stakeholders in using
system data, measurement, and technology to help identify, manage and prevent
gambling risk and harm among their customers.
Dr. Schellinck, PhD, is an Adjunct Professor in the Faculty of Graduate Studies and
Department of Psychology and Neuroscience at Dalhousie University. Her research
is primarily focused on learning and memory in animal models of neurodegenerative
Dr. Michael Bliemel is an associate professor of Management Information Systems
at Dalhousie University in Halifax, NS. He completed his PhD at McMaster
University in Management Science/Systems, specializing in the quantitative
modeling of consumer behaviour with health information systems. His current
research interests include the strategic management of information systems and
innovation in organizations, and business intelligence applications.
... The assessment instruments used in the retrieved prevalence studies on gambling have been Problem Gambling Severity Index (PGSI) (Ferris & Wynne, 2001), South Oaks Gambling Screen (SOGS) (Lesieur & Blume, 1987), National Opinion Research Center DSM Screen for Gambling Problems (NODS) (Gerstein et al., 1999), the American Psychiatric Association's Diagnostic criteria for pathological gambling (DSM-IV) (APA, 1994), and the Focal Adult Gambling Screen (FLAGS) (Schellinck et al., 2015). These tools are used in different contexts (clinical or non-clinical, offline or online) and with different aims (diagnosis or research). ...
... According to the SOGS, probable pathological gambling = 5 + points; according to the NODS, pathological gambling = 5 + points. In this meta-analysis, low/early/intermediate risk gambling is not examined since instruments that consider it (FLAGS and PGSI) specify that low/early/intermediate risk gamblers do not have any indications of impaired control or harm and are characterized as pre-harm risk groups (Ferris & Wynne, 2001;Schellinck et al., 2015). ...
Full-text available
Gambling is widely considered a socially acceptable form of recreation. However, for a small minority of individuals, it can become both addictive and problematic with severe adverse consequences. The aim of this systematic review and meta-analysis is to provide an overview of prevalence studies published between 2016 and the first quarter of 2022 and an updated estimate of problem gambling in the general adult population. A systematic review and a meta-analysis were carried out using academic databases, Internet, and governmental websites. Following this search and utilizing exclusion criteria, 23 studies on adult gambling prevalence were identified, distinguishing between moderate risk/at risk gambling and problem/pathological gambling. This study found a prevalence of moderate risk/at risk gambling to be 2.43% and of problem/pathological gambling to be 1.29% in the adult population. As difficult as it may be to compare studies due to different methodological procedures, cutoffs, and time frames, the present meta-analysis highlights the variations of prevalence across different countries, giving due consideration to the differences between levels of risk and severity. This work intends to provide a starting point for policymakers and academics to fill the gaps on gambling research—more specifically in some countries where the lack of research in this field is evident—and to study the effectiveness of policies implemented to mitigate gambling harm.
... Behaviour tracking technology is likely to be most effective if provided in the context of a binding, universal pre-commitment system that (a) requires gamblers to set a loss limit before commencing a session, and (b) prevents further gambling once the loss limit is reached (Williams et al. 2007). In addition, algorithms could be applied to detect emerging gambling patterns indicative of harm (Schellinck et al. 2015) and, once detected, automated messages could alert gamblers to the potential for harm and ways to avoid this. The introduction of these systems could achieve significant improvements to current consumer protection and harm minimisation efforts. ...
Responsible Gambling Codes of Conduct (CoC) are used around the world to describe electronic gambling machine (EGM) operator commitments to reducing harm from gambling. In addition to the provision of passive product information and warnings, CoC describe how venues should assist EGM users displaying signs of problematic gambling. The focus in this paper is on venue adherence to the active strategies described in these documents relating to supporting ‘responsible gambling’ and discouraging harmful, intensive and extended gambling. The paper triangulates data from aspirational statements by EGM operators published in CoC documents; structured, unannounced observations by the research team in 11 EGM venues; and interviews and focus groups conducted with 40 gamblers and 20 professionals in Melbourne, Australia. Results showed only isolated evidence of supportive interactions between staff and gamblers to address gambling harm. The weight of evidence demonstrated that venues often fail to respond to signs of gambling problems and instead encourage continued gambling in contradiction of their CoC responsibilities. Signs of gambling problems are a normalised feature of EGM use in these venues. To genuinely address this public health and public policy challenge, improved consumer protection for gamblers may be achieved through legislation requiring venues to respond to signs of gambling problems. This may include a range of measures such as banning food and beverage service at machines and limiting withdrawals of cash by gamblers, as well as using behavioural tracking algorithms to identify problematic gambling patterns and binding universal pre-commitment systems to complement supportive interventions by venue staff.
... If possible, indicators should be used in conjunction with algorithmic data generated from the analysis of player behavior. As Schellinck, Schrans, Schellinck, and Bliemel (2015) has pointed out, it is possible to use real-time system data to identify the patterns of behavior that are statistically more indicative of problem gamblers (as based on independent validation using standardized measures). Such data could be used to select certain players for more detailed behavioral observation in the venue. ...
Full-text available
Background and aims In many jurisdictions, where gambling services are provided, regulatory codes require gambling operators to apply a duty of care toward patrons. A common feature of these provisions is some expectation that venue staff identify and assist patrons who might be experiencing problems with their gambling. The effectiveness of such measures is, however, predicated on the assumption that there are reliable and observable indicators that might be used to allow problem gamblers to be distinguished from other gamblers. Methods In this study, we consolidate the findings from two large Australian studies (n = 505 and n = 680) of regular gamblers that were designed to identify reliable and useful indicators for identifying problem gambling in venues. Results It was found that problem gamblers are much more likely to report potentially visible emotional reactions, unusual social behaviors, and very intense or frenetic gambling behavior. Discussion and conclusions This study shows that there are a range of indicators that could potentially be used to identify people experiencing problems in venues, but that decisions are most likely to be accurate if based on an accumulation of a diverse range of indicators.
Full-text available
We examine the manner in which the population prevalence of disordered gambling has usually been estimated, on the basis of surveys that suffer from a potential sample selection bias. General population surveys screen respondents using seemingly innocuous “trigger,” “gateway” or “diagnostic stem” questions, applied before they ask the actual questions about gambling behavior and attitudes. Modeling the latent sample selection behavior generated by these trigger questions using up-to-date econometrics for sample selection bias correction leads to dramatically different inferences about population prevalence and comorbidities with other psychiatric disorders. The population prevalence of problem or pathological gambling in the United States is inferred to be 7.7%, rather than 1.3% when this behavioral response is ignored. Comorbidities are inferred to be much smaller than the received wisdom, particularly when considering the marginal association with other mental health problems rather than the total association. The issues identified here apply, in principle, to every psychiatric disorder covered by standard mental health surveys, and not just gambling disorder. We discuss ways in which these behavioral biases can be mitigated in future surveys.
Full-text available
We study Danish adult gambling behavior with an emphasis on discovering patterns relevant to public health forecasting and economic welfare assessment of policy. Methodological innovations include measurement of formative in addition to reflective constructs, estimation of prospective risk for developing gambling disorder rather than risk of being falsely negatively diagnosed, analysis with attention to sample weights and correction for sample selection bias, estimation of the impact of trigger questions on prevalence estimates and sample characteristics, and distinguishing between total and marginal effects of risk-indicating factors. The most significant novelty in our design is that nobody was excluded on the basis of their response to a 'trigger' or 'gateway' question about previous gambling history. Our sample consists of 8405 adult Danes. We administered the Focal Adult Gambling Screen to all subjects and estimate prospective risk for disordered gambling. We find that 87.6% of the population is indicated for no detectable risk, 5.4% is indicated for early risk, 1.7% is indicated for intermediate risk, 2.6% is indicated for advanced risk, and 2.6% is indicated for disordered gambling. Correcting for sample weights and controlling for sample selection has a significant effect on prevalence rates. Although these estimates of the 'at risk' fraction of the population are significantly higher than conventionally reported, we infer a significant decrease in overall prevalence rates of detectable risk with these corrections, since gambling behavior is positively correlated with the decision to participate in gambling surveys. We also find that imposing a threshold gambling history leads to underestimation of the prevalence of gambling problems.
Full-text available
This is the first of two papers describing the development of the FocaL Adult Gambling Screen for Electronic Gambling Machine players (FLAGS-EGM). FLAGS-EGM is a measurement approach for identifying gambling risk, a tool that incorporates separate reflective and formative constructs into a single instrument. A set of statements was developed that captured ten constructs associated with gambling risk or which were considered components of problem gambling. Following completion of focus groups with regular slot players, a survey with the reduced set of statements was then administered to a sample of 374 casino slot players in Ontario, Canada. Nine of the proposed constructs passed tests for reliability and validity (Risky Cognitions Beliefs, Risky Cognitions Motives, Preoccupation Desire, Risky Practices Earlier, Risky Practices Later, Impaired Control Continue a Session, Impaired Control Begin a Session, Negative Consequences, and Persistence). A tenth construct (Preoccupation Obsession) requires further development through the addition of improved statements. © 2015 Centre for Addiction and Mental Health. All rights reserved.
Full-text available
The tetrachoric correlation describes the linear relation between two continuous variables that have each been measured on a dichotomous scale. The treatment of the point estimate, standard error, interval estimate, and sample size requirement for the tetrachoric correlation is cursory and incomplete in modern psychometric and behavioral statistics texts. A new and simple method of accurately approximating the tetrachoric correlation is introduced. The tetrachoric approximation is then used to derive a simple standard error, confidence interval, and sample size planning formula. The new confidence interval is shown to perform far better than the confidence interval computed by SAS. A method to improve the SAS confidence interval is proposed. All of the new results are computationally simple and are ideally suited for textbook and classroom presentations.
Full-text available
Structural equation modeling (SEM) has become a quasi-standard in marketing and management research when it comes to analyzing the cause-effect relations between latent constructs. For most researchers, SEM is equivalent to carrying out covariance-based SEM (CB-SEM). While marketing researchers have a basic understanding of CB-SEM, most of them are only barely familiar with the other useful approach to SEM-partial least squares SEM (PLS-SEM). The current paper reviews PLS-SEM and its algorithm, and provides an overview of when it can be most appropriately applied, indicating its potential and limitations for future research. The authors conclude that PLS-SEM path modeling, if appropriately applied, is indeed a "silver bullet" for estimating causal models in many theoretical models and empirical data situations.
Full-text available
This research examines the area of online consumer health information retrieval as a field of study that pertains to consumers' use of the Internet to locate and evaluate health related information for the purposes of self education and collection of facts to enable informed decision making. A research model exploring the antecedents of consumer satisfaction with online health information retrieval is developed using constructs from the Information Systems and Human Computer Interaction bodies of literature. This model is quantitatively validated using structural equation modeling techniques. The findings of this research provide evidence that content quality, technical adequacy and trust explain a large proportion of the variance in consumer satisfaction with online health information retrieval for consumers. Appearance and specific content on Web sites played a much smaller role in predicting consumer satisfaction with online health information retrieval.
A critical element in the evolution of a fundamental body of knowledge in marketing, as well as for improved marketing practice, is the development of better measures of the variables with which marketers work. In this article an approach is outlined by which this goal can be achieved and portions of the approach are illustrated in terms of a job satisfaction measure.
Several papers have appeared criticizing the kappa coefficient because of its tendency to fluctuate with sample base rates. The importance of these criticisms is difficult to evaluate because they are presented with regard to a highly specific model of diagnostic decision making. In this article, diagnostic decision making is viewed as a special case of signal detection theory. Each diagnostic process is characterized by a function that relates the probability of a case receiving a positive diagnosis to the severity or salience of symptoms. The shape of this diagnosability curve greatly affects the value of kappa obtained in a study of interrater reliability, how it changes in response to variation in the base rates, and how closely it corresponds to the validity of diagnostic decisions. The common practice of evaluating a diagnostic procedure, when criterion diagnoses for comparison are unavailable, on the basis of the magnitude of the kappa coefficient observed in a reliability study is questionable. New methods for measuring interrater agreement are necessary, and possible directions for research in this area are discussed. (PsycINFO Database Record (c) 2012 APA, all rights reserved)
The Alcohol Use Disorders Identification Test (AUDIT) is a 10–item questionnaire designed by the World Health Organization to screen for hazardous alcohol intake in primary health care settings. In this longitudinal study we examine its performance in predicting alcohol-related harm over the full range of its scores using receiver operating characteristic analyses. Three hundred and thirty ambulatory care patients were interviewed using a detailed assessment schedule which included the AUDIT questions. After 2-3 years, subjects were reviewed and their experience of alcohol-related medical and social harm assessed by interview and perusal of medical records. A UDIT was a good predictor of both alcohol-related social and medical problems. Cut-off points of 7-8 maximized discrimination in the prediction of trauma and hypertension. Higher cut-offs (12 and 22) provided better discrimination in the prediction of alcohol-related social problems and of liver disease or gastrointestinal bleeding, but high specificity was offset by reduced sensitivity. We conclude that the recommended cut-off score of eight is a reasonable approximation to the optimal for a variety of endpoints.